Antihypertensive Drug Therapies and the Risk of Ischemic Stroke

ORIGINAL INVESTIGATION Antihypertensive Drug Therapies and the Risk of Ischemic Stroke

Olaf H. Klungel, PharmD, PhD; Susan R. Heckbert, MD, PhD; W. T. Longstreth, Jr, MD, MPH; Curt D. Furberg, MD, PhD; Robert C. Kaplan, MS, PhD; Nicholas L. Smith, MPH, PhD; Rozenn N. Lemaitre, MPH, PhD; Hubert G. M. Leufkens, PharmD, PhD; Anthonius de Boer, MD, PhD; Bruce M. Psaty, MD, PhD

Background: The relative effectiveness of various an- mic stroke was higher among users of a ␤-blocker (risk tihypertensive drugs with regard to the reduction of stroke ratio [RR], 2.03; 95% confidence interval [CI], 1.05- incidence remains uncertain. 3.94), calcium channel blocker (RR, 2.30; 95% CI, 1.16- 4.56), or angiotensin-converting enzyme inhibitor (RR, Objective: To assess the association between first is- 2.79; 95% CI, 1.47-5.27) than among users of a thiazide chemic stroke and use of antihypertensive drugs. diuretic alone. Among 673 single-drug users with a history of cardiovascular disease, the RRs were 1.22 (95% Methods: A population-based case-control study was CI, 0.63-2.35), 1.18 (95% CI, 0.59-2.33), and 1.45 (95% performed among enrollees of the Group Health Coop- CI, 0.70-3.02) among users of a ␤-blocker, calcium chan- erative of Puget Sound. Case patients included pharma- nel blocker, and angiotensin-converting enzyme inhibi- cologically treated hypertensive patients who sustained tor, respectively, compared with users of a thiazide di- a first ischemic stroke (fatal or nonfatal; n=380) be- uretic alone. tween July 1, 1989, and December 31, 1996. Control subjects were a random sample of treated hypertensive en- Conclusions: In this study of pharmacologically treated rollees without a history of a stroke (n=2790). Medical hypertensive patients, antihypertensive drug regimens that record review and a telephone interview of consenting did not include a thiazide diuretic were associated with survivors were used to collect information on risk fac- an increased risk of ischemic stroke compared with regi- tors for stroke. Computerized pharmacy records were used mens that did include a thiazide. These results support to assess antihypertensive drug use. the use of thiazide diuretics as first-line antihypertensive agents. Results: Among 1237 single-drug users with no history of cardiovascular disease, the adjusted risk of ische- Arch Intern Med. 2001;161:37-43

HE PRIMARY purpose of the less effective in preventing coronary pharmacological treat- heart disease,9 especially in the elderly.10 From the Cardiovascular ment of hypertension is to The Systolic Hypertension in Europe Health Research Unit, prevent major cardiovas- Trial has demonstrated that compared Departments of Epidemiology cular complications such with placebo, the calcium channel (Drs Klungel, Heckbert, Tas stroke. The 4 most widely used antihy- blocker nitrendipine reduces the inci- Longstreth, Kaplan, Smith, pertensive drug classes include diuretics, dence of stroke in older adults with iso- and Psaty), Medicine ␤-blockers, calcium channel blockers, lated systolic hypertension.11 (Drs Longstreth, Lemaitre, and and angiotensin-converting enzyme Thesefindingsfromplacebo-controlled Psaty), Neurology ␤ (Dr Longstreth), and Health (ACE) inhibitors. Randomized clinical trials suggest that -blockers, diuretics, and Services (Dr Psaty), University trials (RCTs) have not demonstrated calcium channel blockers are more effective of Washington, Seattle; the major differences between these antihy- than placebo in the primary prevention of Department of pertensive drug classes with regard to stroke. Less consistent are the findings from Pharmacoepidemiology and lowering of blood pressure,1-4 quality of individual RCTs with separate treatment Pharmacotherapy, Utrecht life,3,5 or regression of left ventricular arms for ␤-blockers and diuretics.12-15 Al- Institute of Pharmaceutical mass.6-8 Results of recent meta-analyses of though findings from the International Pro- Sciences, Utrecht, the antihypertensive drug treatment com- spective Primary Prevention Study in Hy- Netherlands (Drs Klungel, pared with placebo suggest that low-dose pertension trial14 suggested similar reduc- Leufkens, and de Boer); and the ␤ Department of Public Health diuretic therapy is effective in reducing tionsinstrokeincidencewith -blockersand Sciences, Wake Forest the risk of stroke, coronary heart disease, diuretics, the Heart Attack Primary Preven- 15 University School of Medicine, congestive heart failure, and total mortal- tion in Hypertension (HAPPHY) and the Winston-Salem, NC ity, whereas ␤-blockers prevent stroke MedicalResearchCounciltrials12,13 observed (Dr Furberg). and congestive heart failure but seem a larger reduction of stroke incidence with

(REPRINTED) ARCH INTERN MED/ VOL 161, JAN 8, 2001 WWW.ARCHINTERNMED.COM 37

©2001 American Medical Association. All rights reserved. SUBJECTS AND METHODS available in the medical record. The pharmacy data were searched for antihypertensive drug prescriptions imme- SETTING diately preceding the index date. When a subject who was assumed to be at least 80% compliant received The setting was the Group Health Cooperative of Puget enough pills to last until the index date, that person was Sound (GHC), a large staff-model health maintenance or- classified as a potential current user on that date. This ganization with more than 400000 members in western process was repeated to assess use at 30 and 60 days Washington State. before the index date. A current user of antihypertensive drugs was defined as a user for at least 30 days before SUBJECTS and on the index date. This definition excludes recent starters or switchers of antihypertensive drug therapies, Case patients included GHC enrollees, aged 30 to 79 years, whose drug course was first prescribed or changed who were treated pharmacologically for hypertension and who within 30 days of the index date. sustained an incident fatal or nonfatal ischemic stroke be- A subject was considered pharmacologically treated for tween July 1, 1989, and December 31, 1996. Potential ische- hypertension when a recording of antihypertensive drug use mic stroke cases were identified from computerized GHC hos- for the indication of hypertension was present in the medi- pital discharge diagnoses, Washington State death files, and cal records and when the subject was classified as a current the billing records for GHC enrollees who received medical antihypertensive drug user at the index date according to data care or services from non-GHC providers. Diagnostic crite- available in the computerized pharmacy database. ria for ischemic stroke were adopted from the Cardiovascu- lar Health Study.18 These criteria included (1) rapid onset of STATISTICAL ANALYSES neurologic deficit or subarachnoid hemorrhage, (2) deficit persisting for longer than 24 hours unless computed tomog- Complete data were uniformly available from the medical raphy or magnetic resonance imaging show evidence of per- records for case-control status and medical conditions such manent damage, and (3) no underlying brain trauma, tu- as pharmacological treatment of hypertension, angina, and mor, or infection to cause symptoms. Control subjects were diabetes. In preliminary analysis of demographic and be- obtained from a companion study of risk factors for myocar- havioral risk factor data, such as smoking, physical activ- dial infarction at GHC.19 Controls were a randomly selected ity, race, marital status, and educational level, the agree- sample of these GHC enrollees who were treated pharmaco- ment between medical record and self-reported measures logically for hypertension and were frequency matched to the (telephone interview) was good to excellent. Self- myocardial infarction cases by sex, age (within decade), and reported data, if available, were used for these variables; if calendar year. not, then data from the medical record were used. Data were Each subject was assigned an index date. For the cases, missing on smoking (1.1% of subjects), physical activity the index date was the date of the stroke; for controls, the (8.0%), race (2.4%), educational level (27.0%), total cho- index date was a randomly selected date within the calendar lesterol level (4.8%), duration of treatment for hyperten- year for which they had been selected as controls. In most sion (10.5%), and pretreatment blood pressure (30.1%). strata based on sex, 10-year age categories, and index year, We used an approximate Bayesian bootstrap method to im- the control-case ratio was larger than 3:1. We excluded cases pute missing values. This multiple imputation method is and controls (1) who were enrollees for less than 1 year or a modification of the hot-deck method and takes account who had fewer than 4 visits before their index dates, (2) who of the imputation variability.20 In sensitivity analysis, the had had a previous stroke, (3) who had a diagnosis of con- results using multiple imputed data were similar to those gestive heart failure, (4) who did not have a diagnosis of hy- in the analysis limited to subjects with complete data. pertension in their medical record, and (5) whose stroke was All statistical tests were 2-tailed. We used stratification a complication of a procedure or surgery. A history of stroke and logistic regression to control for potential confounders was assessed by medical record review. Subjects with a his- of the association between antihypertensive drug therapy and tory of congestive heart failure were excluded because of con- ischemic stroke, and odds ratios to estimate the relative risk cern about confounding by the indication of congestive heart (RR). Data were analyzed using commercially available soft- failure for ACE inhibitors. ware (SAS, Version 6.12; SAS Institute, Cary, NC). The association of ischemic stroke with antihyperten- DATA COLLECTION AND DEFINITIONS sive drug therapies was assessed separately for subjects with and without clinical cardiovascular disease (CVD). We de- Information on demographics, health habits, cardiovascu- fined CVD as possible, probable, or definite diagnoses of lar risk factors, and comorbidities were abstracted from angina, claudication, cardiac arrhythmias (including atrial medical records or obtained from a telephone interview of or ventricular arrhythmia) or history of myocardial infarc- consenting survivors. Abstraction of the information from tion, transient ischemic attack, coronary angioplasty, coro- the medical records was performed by trained research nary bypass surgery, or carotid endarterectomy. First, we assistants who were aware of case-control status but compared single-drug users of one of the major antihyper- unaware of the purpose of the study. tensive drug classes and users of major 2-drug combina- The GHC computerized pharmacy database was used tions, with single-drug users of benzothiadiazide (thia- to assess antihypertensive drug use. The pharmacy zide) diuretics as the reference group. Second, we compared records contain information about the type, dose, and antihypertensive drug regimens that did not include a thia- quantity of drug dispensed; the prescription fill date; and zide diuretic with regimens that included a thiazide di- dosing instructions. When dosing instructions were miss- uretic, among single-drug users and users of 2 antihyper- ing from the pharmacy database, we used the instructions tensive drugs from different drug classes.

(REPRINTED) ARCH INTERN MED/ VOL 161, JAN 8, 2001 WWW.ARCHINTERNMED.COM 38

©2001 American Medical Association. All rights reserved. diuretic therapy than ␤-blocker therapy. The Captopril Prevention Project trial16 reported an increased risk of Table 1. Characteristics of Ischemic stroke in subjects receiving captopril compared with Stroke Cases and Controls* ␤-blockers or diuretics. However, the recently com- pleted Swedish Trial in Old Patients with Hyperten- Controls Cases 17 sion-2 Study demonstrated that in hypertensive pa- Characteristic No.* Data† No.* Data† tients older than 70 years, calcium antagonists and ACE Age, y 2790 65.6 380 70.3‡ inhibitors apparently did not differ from conventional Females, % 2790 33.3 380 53.7‡ therapy (␤-blockers or diuretics) with regard to the re- Most recent blood duction of the incidence of cardiovascular morbidity and pressure mortality, including stroke. The Captopril Prevention Systolic, mm Hg 2789 142.7 380 150.8‡ Project and the Swedish Trial in Old Patients with Hy- Diastolic, mm Hg 2789 83.5 380 83.5 pertension-2 trials were not designed to distinguish be- Pretreatment blood tween ␤-blockers and diuretics. pressure Systolic, mm Hg 1968 162.9 247 169.3‡ To assess the association between antihypertensive Diastolic, mm Hg 1968 99.7 247 100.1 drug therapy and incident ischemic stroke, we con- Duration of treated 2513 11.2 324 13.3‡ ducted a population-based case-control study among phar- hypertension, mo macologically treated hypertensive patients. No. of antihypertensive 2790 1.4 380 1.5‡ drugs RESULTS Cholesterol, mmol/L 2653 6.03 (232.8) 364 6.28 (242.6) (mg/dL) 2 During the study period, 611 treated hypertensive pa- Body mass index, kg/m 2726 28.6 365 28.6 tients were hospitalized for or died out of the hospital of a Current smoking, % 2760 12.7 376 15.4 Sedentary, % 2582 20.2 333 28.8‡ first ischemic stroke. We also identified 3505 population- Less than high school 2095 13.4 218 16.1 based controls who were eligible. We excluded 92 cases education, % and 199 controls with congestive heart failure, 113 cases Married, % 2790 75.4 380 61.8‡ and 453 controls who were not sufficiently compliant with White, % 2731 89.6 364 90.1 their antihypertensive drug regimens to be classified as cur- Diabetes, % 2790 11.1 380 30.5‡ rent users, and 66 cases and 181 controls who recently Any cardiovascular 2790 37.5 380 60.3‡ started or switched any antihypertensive drug therapy. In disease, % total, 231 cases and 715 controls were excluded for 1 or History of 2790 15.2 380 25.0‡ myocardial more of these reasons. We included 380 ischemic stroke infarction cases (21 fatal and 359 nonfatal) and 2790 controls who History of 2790 5.6 380 19.2‡ were treated pharmacologically for hypertension. transient The clinical characteristics of cases and controls are ischemic attack summarized in Table 1. Compared with controls, cases Atrial fibrillation 2790 5.0 380 14.5‡ were older and more likely to be men and had a higher Angina 2790 19.0 380 27.9‡ systolic blood pressure at treatment and before treat- Cardiovascular 2790 9.2 380 12.9‡ ment. A number of risk factors for ischemic stroke were procedure more common among cases than controls. *Indicates numbers for whom data were available. Because of concern about confounding by indica- †Data are given as means unless otherwise indicated. tion, we compared characteristics of controls who were us- ‡PϽ.05, cases vs controls. ers of thiazide diuretics with controls who used other antihypertensive drugs separately for subjects with and without a history of CVD (Table 2). Among controls with no his- higher among users of single-drug therapy with tory of CVD, users of thiazide diuretics were slightly older ␤-blockers, calcium channel blockers, or ACE inhibi- and had a slightly longer history of treated hypertension, tors than among users of a thiazide diuretic alone and fewer had diabetes than users of nonthiazides. Other (Table 3). The use of ␤-blockers, calcium channel risk factors for stroke were similar for users of thiazide di- blockers, or ACE inhibitors in combination with a thia- uretics compared with users of other antihypertensive drugs. zide diuretic was not significantly associated with an in- Among control subjects with a history of CVD, users of thia- creased risk of ischemic stroke compared with the use zide diuretics were slightly younger, had a lower treated of a thiazide diuretic alone. The use of any 2 antihyper- systolic blood pressure, and were less likely to have dia- tensive drugs not including a thiazide diuretic was as- betes, angina, or a history of cardiovascular procedures than sociated with a 2.48-fold increase in the risk of ischemic users of an antihypertensive drug regimen that did not in- stroke compared with the use of a thiazide diuretic clude a thiazide diuretic. alone. Among 186 cases and 912 controls with a history of CVD, the use of calcium channel blockers in combi- STROKE RISK ASSOCIATED WITH INDIVIDUAL nation with a thiazide diuretic or a nonthiazide 2-drug ANTIHYPERTENSIVE DRUG CLASSES VS combination were each associated with an increased THIAZIDE DIURETICS ALONE risk of ischemic stroke, compared with users of thiazide Among 127 cases and 1566 controls free of CVD, the diuretics alone (Table 3). Use of other single antihyper- adjusted risk of ischemic stroke was 2.03- to 2.79-fold tensive drugs or 2-drug combinations, compared with

(REPRINTED) ARCH INTERN MED/ VOL 161, JAN 8, 2001 WWW.ARCHINTERNMED.COM 39

CVD Absent CVD Present

Thiazide (n = 840) No Thiazide (n = 904) Thiazide (n = 397) No Thiazide (n = 649) Most recent blood pressure Systolic, mm Hg 141.2 142.7 141.4 145.2† Diastolic, mm Hg 84.2 84.7 81.3 81.2 Pretreatment blood pressure Systolic, mm Hg 162.2 163.0 164.8 163.4 Diastolic, mm Hg 100.4 99.6 98.8 96.8† Duration of treated hypertension, y 11.1 9.8† 13.9 12.3† Age, y‡ 65.8 61.6† 69.2 70.2† Total cholesterol, mmol/L (mg/dL) 5.95 (229.6) 6.05 (233.5) 6.12 (236.2) 6.24 (241.1) Body mass index, kg/m2 29.1 28.8 27.9 28.1 Current smoking, % 15.2 12.2 12.9 12.5 Diabetes,% 8.1 12.1† 8.1 16.8† History of myocardial infarction, % NA NA 37.1 40.5 History of transient ischemic attack, % NA NA 13.3 14.9 Angina, % NA NA 41.8 54.8† Claudication, % NA NA 14.6 13.9 Episode of congestive heart failure, % NA NA 8.0 8.2 Atrial fibrillation, % NA NA 12.2 14.2 Cardiovascular procedure, % NA NA 17.8 24.1†

*Comparisons are among control subjects treated for hypertension, adjusted for age, sex, and calender year. Thiazide indicates benzothiadiazide diuretics; CVD, cardiovascular disease; and NA, not applicable. Data are given as means unless otherwise indicated. †PϽ.05. ‡Adjusted for sex and calender year.

Table 3. Adjusted Association Between Ischemic Stroke and Antihypertensive Drug Therapies*

CVD Absent CVD Present

Drug No. of Cases No. of Controls Adjusted OR (95% CI) No. of Cases No. of Controls Adjusted OR (95% CI) Thiazide 22 438 1.00 (Reference) 25 148 1.00 (Reference) ␤-Blocker Alone 20 263 2.03 (1.05-3.94) 27 176 1.22 (0.63-2.35) With thiazide 11 154 1.50 (0.70-3.25) 15 86 1.20 (0.57-2.53) Calcium antagonist Alone 19 190 2.30 (1.16-4.56) 26 151 1.18 (0.59-2.33) With thiazide 7 63 1.88 (0.75-4.75) 19 58 2.48 (1.17-5.29) ACE inhibitor Alone 27 258 2.79 (1.47-5.27) 22 98 1.45 (0.70-3.02) With thiazide 5 83 1.26 (0.44-3.60) 4 33 0.67 (0.20-2.29) Nonthiazide combination 16 117 2.48 (1.20-5.11) 48 162 2.04 (1.09-3.83)

*CVD indicates cardiovascular disease; OR, odds ratio; CI, confidence interval; and ACE, angiotensin-converting enzymes. Among subjects without CVD, the RRs were adjusted for age, sex, calender year, diabetes, total cholesterol level, pretreatment systolic blood pressure, current smoking, and current use of aspirin. Among subjects with CVD, the RRs were also adjusted for history of myocardial infarction, transient ischemic attack, angina, atrial fibrillation, cardiovascular procedure, and current use of hepatic hydroxymethylglutaryl coenzyme reductase inhibitors. Nonthiazide combination includes 2 antihypertensive drugs, neither a thiazide diuretic. the use of thiazide diuretics alone, was not significantly who used a thiazide diuretic (RR, 1.85; 95% confidence associated with an increased risk of ischemic stroke. interval [CI], 1.26-2.71). Among subjects with a history of clinical CVD who did not use a thiazide diuretic, the STROKE RISK ASSOCIATED WITH increased adjusted risk of ischemic stroke compared with NONTHIAZIDE ANTIHYPERTENSIVE users of thiazide diuretics was not statistically signifi- DRUGS VS THIAZIDE DIURETICS cant (RR, 1.25; 95% CI, 0.87-1.80). Additional adjustment for educational level, mari- The second analysis was conducted among 348 cases and tal status, duration of treated hypertension, treated di- 2608 controls who used 1 or 2 antihypertensive drugs astolic blood pressure, pretreatment diastolic blood pres- (Table 4). Among 142 cases and 1652 controls free of sure, body mass index (calculated as weight in kilograms CVD, subjects who did not use a thiazide diuretic had, divided by the square of height in meters, physical ac- after adjustment for potential confounding factors, an in- tivity, glucose and potassium levels, and use of alcohol creased risk of ischemic stroke compared with subjects had trivial effects on the findings.

(REPRINTED) ARCH INTERN MED/ VOL 161, JAN 8, 2001 WWW.ARCHINTERNMED.COM 40

CVD Absent CVD Present

No. of Cases/Controls Adjusted OR (95% CI) No.of Cases/Controls Adjusted OR (95% CI) 1 Drug Thiazide 22/438 1.00 (Reference) 25/148 1.00 (Reference) No thiazide 75/760 2.42 (1.43-4.07) 91/460 1.40 (0.81-2.43) 2 Drugs Thiazide 29/337 1.00 (Reference) 42/186 1.00 (Reference) No thiazide 16/117 1.40 (0.70-2.78) 48/162 1.33 (0.78-2.26) 1or 2 Drugs Thiazide 51/775 1.00 (Reference) 67/334 1.00 (Reference) No thiazide 91/877 1.85 (1.26-2.71) 139/622 1.25 (0.87-1.80)

*Thiazide indicates benzothiadiazide diuretics; CVD, cardiovascular disease; OR, odds ratio; and CI, confidence interval. Among subjects without CVD, the RRs were adjusted for age, sex, calender year, diabetes, total cholesterol level, pretreatment systolic blood pressure, current smoking, and current use of aspirin. Among subjects with CVD, the RRs were also adjusted for history of myocardial infarction, transient ischemic attack, angina, atrial fibrillation, cardiovascular procedure, and current use of hepatic hydroxymethylglutaryl coenzyme reductase inhibitors.

Stratified analysis revealed no effect modification of pared with diuretics (RR, 1.29; 95% CI, 0.84-1.83). the association between ischemic stroke and use of thia- Findings from the Medical Research Council trials zide diuretics in subgroups based on sex, median age (70 clearly suggest a higher risk of stroke with ␤-blocker years), presence of diabetes, smoking status, median pre- therapy than with diuretic therapy in middle-aged treatment blood pressure (diastolic, 100 mm Hg; sys- subjects (RR, 2.28; 95% CI, 1.31-3.96),12 and a nonsig- tolic, 170 mm Hg), median treated blood pressure (di- nificantly higher risk in older subjects (RR, 1.23; 95% astolic, 84 mm Hg; systolic, 152 mm Hg), median total CI, 0.86-1.79).13 Results from the Captopril Preven- serum cholesterol level (6.16 mmol/L [238 mg/dL]), and tion Project trial showed an increased risk of stroke median serum potassium level (4.1 mmol/L). The re- with captopril therapy compared with ␤-blocker or sults were virtually the same for the use of thiazide di- diuretic therapy (RR, 1.25; 95% CI, 1.01-1.55). This uretics below the daily modal dose and above or at the increased risk of stroke may have been due to a failure daily modal dose (25 mg for hydrochlorothiazide, 25 mg of randomization.21 In a clinical trial designed to assess for chlorthalidone, and 5 mg for metolazone) and for the the effects of antihypertensive therapy on carotid ath- use of thiazide diuretics alone or in combination with po- erosclerosis,22 subjects randomized to receive the cal- tassium-sparing agents. cium channel blocker isradipine had a higher rate of stroke events than those randomized to receive hydro- COMMENT chlorothiazide (RR, 2.00; 95% CI, 0.50-7.93). In the Swedish Trial in Old Patients with Hypertension-2 In this population-based case-control study among trial, subjects allocated to receive ACE inhibitors and pharmacologically treated hypertensive patients with- calcium antagonists, respectively, had a similar risk of out clinically recognized CVD, the use of antihyper- stroke compared with subjects allocated to receive tensive drug regimens that did not include a thiazide ␤-blocker or diuretic therapy (respective RRs, 0.90 diuretic was associated with an 85% increased risk of [95% CI, 0.74-1.08] and 0.88 [95% CI, 0.73-1.06]). ischemic stroke compared with the use of an antihy- These findings from RCTs tend to favor thiazide pertensive drug regimen that included a thiazide diuretics over other antihypertensive drug therapies diuretic. Even among users of 2 antihypertensive for reduction of stroke risk and are consistent with the drugs, the use of nonthiazide antihypertensive drug findings from our study. Additional support for a par- regimens was associated with a higher risk of ischemic ticular benefit of thiazide diuretics comes from the stroke (40%). This association persisted after adjust- recent interim analysis of the Antihypertensive and ment for many potential confounding factors, and was Lipid-Lowering Treatment to Prevent Heart Attack consistent across a variety of subgroups. Among sub- Trial.23 Compared with patients who received jects with clinically manifest CVD, this association chlorthalidone, patients treated with the ␣-blocker was less pronounced. doxazosin mesylate had an increased risk of stroke The International Prospective Primary Prevention (RR, 1.19; 95% CI, 1.01-1.40). Study in Hypertension trial, in which ␤-blocker One possible explanation for these findings may lie therapy was compared with non–␤-blocker (mostly in the fact that systolic blood pressure is more strongly diuretic) therapy, showed no significant difference in associated with the occurrence of stroke than diastolic stroke incidence between the treatment groups (RR, blood pressure,24 and that thiazide diuretics may be more 0.97; 95% CI, 0.64-1.47). However, this trial may not effective in lowering systolic blood pressure than other allow a valid comparison of stroke risk because 67% of antihypertensive drugs, whereas the effect of thiazide di- the patients allocated to ␤-blocker therapy also uretics on diastolic blood pressure is similar to that of received a diuretic. The HAPPHY trial found a nonsig- other major antihypertensive drug classes.3,4,12,13,16,22,23 Sys- nificantly higher risk of stroke for ␤-blockers com- tolic blood pressure during treatment was also slightly

(REPRINTED) ARCH INTERN MED/ VOL 161, JAN 8, 2001 WWW.ARCHINTERNMED.COM 41

©2001 American Medical Association. All rights reserved. lower among users of thiazide diuretics in this study. Ad- though the mechanism is not clear, the findings are con- justment for systolic blood pressure during treatment had sistent with those of previous RCTs. Ongoing large- little effect on our results. For example, among those re- scale clinical trials should help clarify this issue.27 The ceiving monotherapy, the risk of ischemic stroke asso- Sixth Report of the Joint National Committee on Pre- ciated with not using compared with using a thiazide di- vention, Detection, Evaluation, and Treatment of High uretic decreased from 2.42 to 2.30. However, we were Blood Pressure currently recommends diuretics and not able to test this hypothesis adequately, because many ␤-blockers.28 In the absence of additional clinical trial evi- of the current users of thiazide diuretics were not using dence, the results of our study support the use of thia- a thiazide diuretic at the time when their blood pressure zide diuretics as first-line antihypertensive agents. during treatment was recorded. A biological mechanism independent of blood pres- Accepted for publication July 10, 2000. sure cannot be excluded, as suggested by a post hoc analy- The research reported in this article was supported in sis of the Swedish Trial in Old Patients with Hypertension part by grants HL40628, HL43201, and HL60739 from the trial in which two thirds of the actively treated patients re- National Heart, Lung, and Blood Institute, Bethesda, Md. ceived a ␤-blocker and a thiazide diuretic.25 After match- Dr Psaty is a Merck/SER Clinical Epidemiology Fellow, spon- ing on achieved blood pressure and controlling for initial sored by the Merck Co Foundation, Rahway, NJ, and the blood pressure, the subjects receiving active treatment had Society for Epidemiologic Research, Baltimore, Md. Dr a 42% decreased risk of stroke compared with those receiv- Kaplan was a Howard Hughes Medical Institute Predoc- ing placebo (RR, 0.58; 95% CI, 0.35-0.98), suggesting a non– toral Fellow when this work was conducted. blood pressure mediated benefit of antihypertensive therapy Corresponding author and reprints: Olaf H. Klungel, with ␤-blockers or diuretics beyond that achieved by merely PharmD, PhD, Department of Pharmacoepidemiology and lowering blood pressure. Moreover, recently it was dem- Pharmacotherapy, Utrecht Institute of Pharmaceutical Sci- onstrated that diuretics may have an additional therapeu- ences, Sorbonnelann 16, 3584 CA Utrecht, the Nether- tic advantage by restoring nocturnal blood pressure decline lands (e-mail: [email protected]). in patients with sodium-sensitive hypertension.26 The strengths of this observational study are the use REFERENCES of population-based case-control subjects, the complete- ness of case identification, the comparable ascertainment 1. Philipp T, Anlauf M, Distler A, Holzgreve H, Michaelis J, Wellek S. Randomised, of potential confounding factors, and the use of phar- double blind, multicentre comparison of hydrochlorothiazide, atenolol, nitren- macy records to assess antihypertensive drug use in a com- dipine, and enalapril in antihypertensive treatment: results of the HANE study: parable and unbiased fashion for cases and controls. We HANE Trial Research Group. BMJ. 1997;315:154-159. used restriction, stratification, and multivariate adjust- 2. Applegate WB, Phillips HL, Schnaper H, et al. A randomized controlled trial of the effects of three antihypertensive agents on blood pressure control and qual- ment to minimize the influence of confounding. ity of life in older women. Arch Intern Med. 1991;151:1817-1823. An important limitation of this observational study 3. Neaton JD, Grimm RH Jr, Prineas RJ, et al, for the Treatment of Mild Hyperten- is that antihypertensive drug treatment was not ran- sion Study Research Group. Treatment of Mild Hypertension Study: final re- domly assigned. Physicians and patients selected anti- sults. JAMA. 1993;270:713-724. 4. Materson BJ, Reda DJ, Cushman WC, et al, for the Department of Veterans Af- hypertensive drug therapies, and this may have intro- fairs Cooperative Study Group on Antihypertensive Agents. Single-drug therapy duced bias. Despite adjustment for potential confounding for hypertension in men: a comparison of six antihypertensive agents with pla- factors, residual confounding due to incomplete or in- cebo. N Engl J Med. 1993;328:914-921. accurate measurement of covariates or unmeasured 5. Beto JA, Bansal VK. Quality of life in the treatment of hypertension: a meta- confounders cannot be excluded. analysis of clinical trials. Am J Hypertens. 1992;5:125-133. 6. Dahlof B, Pennert K, Hansson L. Reversal of left ventricular hypertrophy in hy- The preferred design to compare these antihyperten- pertensive patients: a metaanalysis of 109 treatment studies. Am J Hypertens. sive drug therapies in terms of their risk or benefit with re- 1992;5:95-110. gard to cardiovascular outcomes is the controlled RCT. 7. Gottdiener JS, Reda DJ, Massie BM, Materson BJ, Williams DW, Anderson RJ, However, when clinical trial results are lacking or conflict- for the Department of Veterans Affairs Cooperative Study Group on Antihyper- ing, well-designed observational studies can complement tensive Agents. Effect of single-drug therapy on reduction of left ventricular mass in mild to moderate hypertension: comparison of six antihypertensive agents. them. Furthermore, the highly selected nature of partici- Circulation. 1997;95:2007-2014. pants of RCTs and the strict, protocol-driven conditions 8. Liebson PR, Grandits GA, Dianzumba S, et al. Comparison of five antihyperten- under which trials are conducted may limit the generaliz- sive monotherapies and placebo for change in left ventricular mass in patients ability of findings from RCTs to general practice. The com- receiving nutritional-hygienic therapy in the Treatment of Mild Hypertension Study (TOMHS). Circulation. 1995;91:698-706. mon use of a large number of alternative antihypertensive 9. Psaty BM, Smith NL, Siscovick DS, et al. Health outcomes associated with an- drugs makes evaluation of these therapies in an observa- tihypertensive therapies used as first-line agents: a systematic review and meta- tional setting feasible. The high degree of similarity in sev- analysis. JAMA. 1997;277:739-745. eral important clinical characteristics between users of thia- 10. Messerli FH, Grossman E, Goldbourt U. Are ␤-blockers efficacious as first-line zide diuretics and users of other antihypertensive drugs therapy for hypertension in the elderly? a systematic review. JAMA. 1998;279: 1903-1907. among subjects without CVD suggests minimal confound- 11. Staessen JA, Fagard R, Thijs L, et al. Randomised double-blind comparison of ing by those characteristics that were measurable. placebo and active treatment for older patients with isolated systolic hypertension. Lancet. 1997;350:757-764. 12. MRC Working Party. MRC trial of treatment of mild hypertension: principal re- CONCLUSIONS sults. BMJ. 1985;291:97-104. 13. MRC Working Party. MRC trial of treatment of hypertension in older adults: prin- This study suggests a particular benefit of thiazide di- cipal results. BMJ. 1992;304:405-412. uretics in reducing the risk of ischemic stroke. Al- 14. The IPPPSH Collaborative Group. Cardiovascular risk and risk factors in a ran-

(REPRINTED) ARCH INTERN MED/ VOL 161, JAN 8, 2001 WWW.ARCHINTERNMED.COM 42

©2001 American Medical Association. All rights reserved. domized trial of treatment based on the beta-blocker oxprenolol: the Interna- center Isradipine Diuretic Atherosclerosis Study (MIDAS): a randomized con- tional Prospective Primary Prevention Study in Hypertension (IPPPSH). J Hy- trolled trial. JAMA. 1996;276:785-791. pertens. 1985;3:379-392. 23. The ALLHAT Officers and Coordinators for the ALLHAT Collaborative Research 15. Wilhelmsen L, Berglund G, Elmfeldt D, et al. Beta-blockers versus diuretics in hy- Group. Major cardiovascular events in hypertensive patients randomized to dox- pertensive men: main results from the HAPPHY trial.J Hypertens. 1987;5:561-572. azosin vs chlorthalidone: the Antihypertensive and Lipid-Lowering Treatment to 16. Hansson L, Lindholm LH, Niskanen L, et al. Effect of angiotensin-converting- Prevent Heart Attack Trial (ALLHAT). JAMA. 2000;283:1967-1975. enzyme inhibition compared with conventional therapy on cardiovascular mor- 24. Stamler J, Stamler S, Neaton JD. Blood pressure, systolic and diastolic, and car- bidity and mortality in hypertension: the Captopril Prevention Project (CAPPP) diovascular risks: US population data. Arch Intern Med. 1993;153:598-615. randomised trial. Lancet. 1999;353:611-616. 25. Ekbom T, Dahlof B, Hansson L, et al. The stroke preventive effect in elderly hy- 17. Hansson L, Lindholm LH, Ekbom T, et al. Randomised trial of old and new an- pertensives cannot fully be explained by the reduction in office blood pressure: tihypertensive drugs in elderly hypertensive patients: cardiovascular mortality insights from the Swedish Trial in Old Patients with Hypertension (STOP- and morbidity in the Swedish Trial in Old Patients with Hypertension-2 study. Hypertension). Blood Press. 1992;1:168-172. Lancet. 1999;354:1751-1756. 26. Uzu T, Kimura G. Diuretics shift circadian rhythm of blood pressure from non- 18. Price TR, Psaty BM, O’Leary D, Burke G, Gardin J. Assessment of cerebrovascular dipper to dipper in essential hypertension. Circulation. 1999;100:1635-1638. disease in the Cardiovascular Health Study. Ann Epidemiol. 1993;3:504-507. 27. Davis BR, Cutler JA, Gordon DJ, et al, for the ALLHAT Research Group. Ratio- 19. Psaty BM, Heckbert SR, Koepsell TD, et al. The risk of myocardial infarction as- nale and design for the Antihypertensive and Lipid Lowering Treatment to Pre- sociated with antihypertensive drug therapies. JAMA. 1995;274:620-625. vent Heart Attack Trial (ALLHAT). Am J Hypertens. 1996;9:342-360. 20. Rubin DB, Schenker N. Multiple imputation in health-care databases: an over- 28. Joint National Committee on Prevention, Detection, Evaluation, and Treatment view and some applications. Stat Med. 1991;10:585-598. of High Blood Pressure. The Sixth Report of the Joint National Committee on 21. Cutler J. Which drug for treatment of hypertension? Lancet. 1999;353:604-605. Prevention, Detection, Evaluation, and Treatment of High Blood Pressure. Arch 22. Borhani NO, Mercuri M, Borhani PA, et al. Final outcome results of the Multi- Intern Med. 1997;157:2413-2446.

(REPRINTED) ARCH INTERN MED/ VOL 161, JAN 8, 2001 WWW.ARCHINTERNMED.COM 43